DeepMind unveils Genie 4 world model promising more realistic physics and humans

Key Points:

• Google DeepMind has developed Genie 4, an advanced world model that improves on the previous Genie 3 by generating real-time, interactive virtual environments with enhanced realism and native audio, including interactive human speech.
• Genie 4 uses the Gemini 3.8 Flash model to guide simulations, maintaining context across varying environments and allowing users to modify parameters like weather and camera angles mid-simulation for a more dynamic experience.
• The updated model offers more realistic physics and natural character movements, making it potentially valuable for training physical robots and rapid prototyping in creative fields, aligning with DeepMind’s goal of progressing toward artificial general intelligence (AGI).
• Access to Genie 4 is currently limited to a small group of testers, with plans to expand availability through the Gemini API and AI Studio; consumer access will be phased in via Google’s premium AI Ultra plans starting at $100 per month, with a waitlist open for early access.
• Despite its advancements, Genie 4 remains an active research project, and Google has not provided a specific timeline for broader release or detailed pricing, emphasizing that the technology is still evolving.